Halooo, I’m Akilesh.

I am usually building something, breaking something..... or poking at interesting model architectures just to see what happens. Right now I’m deep into inference serving, distributed training and MLOps. I write about what I learn so I don’t forget it lol, and so AI feels a little less confusing for everyone else.

Where latency, memory and batching collide.

Then scroll into the writing

Notes from inside the machine

Things I needed to understand properly.

Ideas, experiments and technical rabbit holes, written down before I forget what made them interesting.

  1. 01Architechural Masterclass in Kimi K3Field Notes
  2. 02What a benchmark cannot tell youResearch
  3. 03Building the smallest version that can teach you somethingBuilds
  4. 04Cache PoisoningResearch
  5. 05Prompt CachingResearch
  6. 06AI Does Multiplication Underneath. So Why Did Older Models Break at School Maths?Research
  7. 07The Smallest Thing in PyTorch Opens Half the GPU StackResearch
  8. 08What Happens When You Put “n” Billion Weights in Your RAMResearch
  9. 09Reading Minds Is the New LoggingResearch
  10. 10The Physics of Prompting: Why Words Don’t Matter, But Structure DoesResearch
Open the full archive (012)

A few loose pages

From the zine stack.

  1. Field Notes · 01Architechural Masterclass in Kimi K3
  2. Research · 02What a benchmark cannot tell you
  3. Builds · 03Building the smallest version that can teach you something